Two RB433AH (v3.23 and 3.27 tested) with two CM9s each, 40mhz channels.
433 #1 (172.16.1.9) has CM9s bonded (bonding1 192.168.1.100), rr, arp link monitoring to 192.168.1.110.
433 #2 (172.16.1.11) has CM9s bonded (bonding1 192.168.1.110), rr, arp link monitoring to 192.168.1.100.
ether1 and bonding1 is bridged on each board with the 172 address bound to the bridge.
A UDP test results in 75mbps thruput, which is balanced across both CM9s, approx 38mbps each. Running TCP results in only 38mbps TOTAL.
Dropping a link results in the remaining link rising to approx 51mbps UDP, and 38mbps TCP.
Changing to 802.3ad yields behaviour similar to active/passive - only one link carries traffic, if it is dropped there is no disruption and the traffic gracefully moves to the other link.
A few months ago we had tested 3.23 with 3 R52s each and successfully ran 210mbs across it, so I am befuddles as to this new behaviour. I’ve also tested with R52s and now achieve the same results as above.
I have also bonded ether 2 and 3 on each board and can only achieve 48mbps on each link, 98mbps total.
Switch chip total bandwidth is 100Mbps so you can’t get more than that bonding two interfaces in the same switch. If you bond ether1 and ether2 on RB433 then you will get almost 200Mbps.
Switch chip is 100, yet the second statement says 200mbps… is that an aggregate number? Or could you restate? I had bonded ether 2 and 3 together, but did not bridge them to any other interface, in hopes that if the bridge were limiting to 100 that it wouldn’t be a problem, with no change in results.
The question posted truly relates to aggregating 2 or 3 wlan interfaces… when bonding came out, I tested and it worked great. I now can not duplicate those results. Everything appears to behave as if they are simply redundant interfaces, not aggregated interfaces.
What mrz is saying is that the actual chip that controls the switch part can’t handle more than 100Mbit/s. I suppose the 433 is similar to the 493 and 450 in the way that it has a chip that allows all ports except ether1 to be grouped into a switch. Since ether1 is not a part of the switch chip, you should be able to get better throughput using a bonding between ether1 and ether2, instead of ether2 and ether3, since they latter are controlled by the same chip.
A UDP test results in 75mbps thruput, which is balanced across both CM9s, approx 38mbps each.
You got 75mbps UDP over two bonded wireless links, so bonding is working.
You will not get the same results with TCP because wireless is not full duplex.